MML-EM Algorithm and Its Application on Mixed Distributions of Geochemical Data

When separating the mixed distributions of element abundance, probability graphs can only make a rough estimate of parameters. To solve this problem, we introduce a method of Minimum Message Length Criterion-Expectation-Maximization Algorithm (MML-EM). Simulation studies have shown that the method has higher accuracy than probability graphs in estimating the parameters of mixed distributions of element abundance. It is applied to dealing with the geochemical data of quartz vein, sampled from tungsten ore in Dajishan, Jiangxi Province. The research shows that the concentrations of W, Ta and Nb are mixed pairs of log-normal distributions, biomodel distributions. Referring to the previous geological results, the conclusion is as follows. For W, the component with high average value represents the high-grade ore formed by hydrothermal filling type in ore-forming processes, and the one with low average value represents the mineralization in other stages, high-value parts of which may indicate that there is a weak disseminated tungsten mineralization at late magmatic stage. For Ta and Nb, the situation is similar. So the method provides a good quantitative tool that can help separate the mixed distributions of geochemical data and explain the multiple geological events.